Liu Yuan, Zhao Songyun, Shang Xingchen, Shen Wei, Du Wenyi, Zhou Ning
Department of General Surgery, Wuxi People's Hospital Affiliated to Nanjing Medical University, Wuxi, China.
Department of Neurosurgery, Wuxi People's Hospital Affiliated to Nanjing Medical University, Wuxi, China.
Front Oncol. 2024 Nov 27;14:1471137. doi: 10.3389/fonc.2024.1471137. eCollection 2024.
Complications and mortality rates following gastrectomy for gastric cancer have improved over recent years; however, complications such as anastomotic leakage (AL) continue to significantly impact both immediate and long-term prognoses. This study aimed to develop a machine learning model to identify preoperative and intraoperative high-risk factors and predict mortality in patients with AL after radical gastrectomy.
For this investigation, 906 patients diagnosed with gastric cancer were enrolled and evaluated, with a comprehensive set of 36 feature variables collected. We employed three distinct machine learning algorithms-extreme gradient boosting (XGBoost), random forest (RF), and k-nearest neighbor (KNN)-to develop our models. To ensure model robustness, we applied k-fold cross-validation for internal validation of the four models and subsequently validated them using independent datasets.
In contrast to the other machine learning models employed in this study, the XGBoost algorithm exhibited superior predictive performance in identifying mortality risk factors for patients with AL across one, three, and five-year intervals. The analysis identified several common risk factors affecting mortality rates at these intervals, including advanced age, hypoproteinemia, a history of anemia and hypertension, prolonged operative time, increased intraoperative bleeding, low intraoperative percutaneous arterial oxygen saturation (SPO) levels, T3 and T4 tumors, tumor lymph node invasion, and tumor peripheral nerve invasion (PNI).
Among the three machine learning models examined in this study, the XGBoost algorithm exhibited superior predictive capabilities concerning the prognosis of patients with AL following gastrectomy. Additionally, the use of machine learning models offers valuable assistance to clinicians in identifying crucial prognostic factors, thereby enhancing personalized patient monitoring and management.
近年来，胃癌胃切除术后的并发症和死亡率有所改善；然而，诸如吻合口漏（AL）等并发症仍然对近期和长期预后产生重大影响。本研究旨在开发一种机器学习模型，以识别术前和术中的高危因素，并预测根治性胃切除术后发生AL患者的死亡率。
在本次调查中，纳入并评估了906例诊断为胃癌的患者，收集了一套全面的36个特征变量。我们采用了三种不同的机器学习算法——极端梯度提升（XGBoost）、随机森林（RF）和k近邻（KNN）——来开发我们的模型。为确保模型的稳健性，我们对四个模型进行了k折交叉验证以进行内部验证，随后使用独立数据集对它们进行验证。
与本研究中使用的其他机器学习模型相比，XGBoost算法在识别1年、3年和5年间隔期内AL患者的死亡风险因素方面表现出卓越的预测性能。分析确定了在这些间隔期影响死亡率的几个常见风险因素，包括高龄、低蛋白血症、贫血和高血压病史、手术时间延长、术中出血增加、术中经皮动脉血氧饱和度（SPO）水平低、T3和T4期肿瘤、肿瘤淋巴结侵犯和肿瘤周围神经侵犯（PNI）。
在本研究中检验的三种机器学习模型中，XGBoost算法在预测胃切除术后AL患者的预后方面表现出卓越的预测能力。此外，机器学习模型的使用为临床医生识别关键预后因素提供了有价值的帮助，从而加强了对患者的个性化监测和管理。
